This includes the usual amount of editorial cleanup, such as
correcting wrong or less-helpful-than-they-could-be examples.
I moved the two tsvector-updating triggers into "9.28 Trigger
Functions", which seems like a better home for them. (I believe
that section didn't exist when this text was originally written.)
Commit 32ff26911 introduced use of rank() into the triggers view to
calculate the spec-mandated action_order column. As written, this
prevents query constraints on the table-name column from being pushed
below the window aggregate step. That's bad for performance of this
typical usage pattern, since the view now has to be evaluated for all
tables not just the one(s) the user wants to see. It's also the cause
of some recent buildfarm failures, in which trying to evaluate the view
rows for triggers in process of being dropped resulted in "cache lookup
failed for function NNN" errors. Those rows aren't of interest to the
test script doing the query, but the filter that would eliminate them
is being applied too late. None of this happened before the rank()
call was there, so it's a regression compared to v10 and before.
We can improve matters by changing the rank() call so that instead of
partitioning by OIDs, it partitions by nspname and relname, casting
those to sql_identifier so that they match the respective view output
columns exactly. The planner has enough intelligence to know that
constraints on partitioning columns are safe to push down, so this
eliminates the performance problem and the regression test failure
risk. We could make the other partitioning columns match view outputs
as well, but it'd be more complicated and the performance benefits
are questionable.
Side note: as this stands, the planner will push down constraints on
event_object_table and trigger_schema, but not on event_object_schema,
because it checks for ressortgroupref matches not expression
equivalence. That might be worth improving someday, but it's not
necessary to fix the immediate concern.
Back-patch to v11 where the rank() call was added. Ordinarily we'd not
change information_schema in released branches, but the test failure has
been seen in v12 and presumably could happen in v11 as well, so we need
to do this to keep the buildfarm happy. The change is harmless so far
as users are concerned. Some might wish to apply it to existing
installations if performance of this type of query is of concern,
but those who don't are no worse off.
I bumped catversion in HEAD as a pro forma matter (there's no
catalog incompatibility that would really require a re-initdb).
Obviously that can't be done in the back branches.
Discussion: https://postgr.es/m/5891.1587594470@sss.pgh.pa.us
DST law changes in Morocco and the Canadian Yukon.
Historical corrections for Shanghai.
The America/Godthab zone is renamed to America/Nuuk to reflect
current English usage; however, the old name remains available as a
compatibility link.
The test is proving to have timing issues when looking at archive status
files on standbys after crash recovery, while other parts of the test
rely on pg_stat_archiver as a wait point to make sure that a given state
of the archiving is reached. The coverage is not heavily impacted by
the removal those extra tests.
Per reports from several buildfarm animals, like crake, piculet,
culicidae and francolin.
Discussion: https://postgr.es/m/20200424005929.GK33034@paquier.xyz
Backpatch-through: 9.5
78ea8b5 has fixed an issue related to the recycling of WAL segments on
standbys depending on archive_mode. However, it has introduced a
regression with the handling of WAL segments ready to be archived during
crash recovery, causing those files to be recycled without getting
archived.
This commit fixes the regression by tracking in shared memory if a live
cluster is either in crash recovery or archive recovery as the handling
of WAL segments ready to be archived is different in both cases (those
WAL segments should not be removed during crash recovery), and by using
this new shared memory state to decide if a segment can be recycled or
not. Previously, it was not possible to know if a cluster was in crash
recovery or archive recovery as the shared state was able to track only
if recovery was happening or not, leading to the problem.
A set of TAP tests is added to close the gap here, making sure that WAL
segments ready to be archived are correctly handled when a cluster is in
archive or crash recovery with archive_mode set to "on" or "always", for
both standby and primary.
Reported-by: Benoît Lobréau
Author: Jehan-Guillaume de Rorthais
Reviewed-by: Kyotaro Horiguchi, Fujii Masao, Michael Paquier
Discussion: https://postgr.es/m/20200331172229.40ee00dc@firost
Backpatch-through: 9.5
In the footsteps of aaf069aa3, remove ACLDEBUG, which was the only
other remaining undocumented symbol in pg_config_manual.h. The fact
that nobody had bothered to document it in seventeen years is a good
clue to its usefulness. In practice, none of the tracing logic it
enabled would be of any value without additional effort.
Discussion: https://postgr.es/m/6631.1587565046@sss.pgh.pa.us
Nobody really uses this stuff, especially not since we created
valgrind-based infrastructure that does the same thing better.
It is thus unsurprising that the generation.c and slab.c versions
were actually broken. Rather than fix 'em, let's just remove 'em.
Alexander Lakhin
Discussion: https://postgr.es/m/8936216c-3492-3f6e-634b-d638fddc5f91@gmail.com
Also rearrange that page a bit for more consistency and less
duplication.
In passing, fix erroneous examples of the results of abbrev(cidr)
in datatype.sgml, and do a bit of copy-editing there.
Function names declared "extern" now use BackupManifest in the name
rather than just Manifest, and data types use backup_manifest
rather than just manifest.
Per note from Michael Paquier.
Discussion: http://postgr.es/m/20200418125713.GG350229@paquier.xyz
In a9c35cf85c I changed ExecMakeTableFunctionResult() to dynamically
allocate the FunctionCallInfo used to call the SRF. Unfortunately I
did not account for the fact that the surrounding memory context has
query lifetime, leading to a leak till the end of the query.
In most cases the leak is fairly inconsequential, but if the
FunctionScan is done many times in the query, the leak can add
up. This happens e.g. if the function scan is on the inner side of a
nested loop, due to a lateral join.
EXPLAIN SELECT sum(f) FROM generate_series(1, 100000000) g(i), generate_series(i, i+1) f;
quickly shows the leak.
Instead of explicitly freeing the FunctionCallInfo it seems better to
make sure all the per-set temporary state in
ExecMakeTableFunctionResult() is cleaned up wholesale. Currently
that's probably just the FunctionCallInfo allocation, but since
there's some initialization work, and since there's already an
appropriate context, this seems like a more robust approach.
Bug: #16112
Reported-By: Ben Cornett
Author: Andres Freund
Reviewed-By: Tom Lane
Discussion: https://postgr.es/m/16112-4448bbf55a404189%40postgresql.org
Backpatch: 12, a9c35cf85c
This commit does:
- get rid of the garbage code for unused --print-parse-wal option.
- add help message for --quiet option into usage().
- fix typo of option name in help message.
Author: Fujii Masao
Reviewed-by: Robert Haas
Discussion: https://postgr.es/m/ff4710f7-2331-4f6b-012e-d76da3275e91@oss.nttdata.com
David Johnston reminded me that the per-point calculations being done
by these operators are equivalent to complex multiplication/division.
(Once I would've recognized that immediately, but it's been too long
since I did any of that sort of math.)
Also put in a footnote mentioning that "rotation" of a box doesn't do
what you might expect, as I'd griped about in the referenced thread.
Discussion: https://postgr.es/m/158110996889.1089.4224139874633222837@wrigleys.postgresql.org
The mask was added by commit 8224de4f, which introduced INCLUDE nbtree
indexes. The status bits really were reserved initially. We now use 2
out of 4 of the bits for additional tuple metadata, though. Rename the
mask to BT_STATUS_OFFSET_MASK.
Also consolidate related nbtree.h code comments about the format of
pivot tuples and posting list tuples.
When estimating the number of pre-sorted groups in cost_incremental_sort
we must not pass Vars with varno 0 to estimate_num_groups, which would
cause failues in find_base_rel. This may happen when sorting output of
set operations, thanks to generate_append_tlist.
Unlike recurse_set_operations we can't easily access the original target
list, so if we find any Vars with varno 0, we fall back to the default
estimate DEFAULT_NUM_DISTINCT.
Reported-by: Justin Pryzby
Discussion: https://postgr.es/m/20200411214639.GK2228%40telsasoft.com
Make header/trailer comments agree with the actual names of some macros.
These seem like legit names in earlier iterations of respective patches
(commit b779168ff "Detect PG_PRINTF_ATTRIBUTE automatically." and
commit 6869b4f25 "Add C++ support to configure.") but the macro had
been renamed out of sync with the header / trailer comment in the final
committed patch.
Even more nitpickily, make the dashed underlines agree with the lengths
of the macro names everyplace. There doesn't seem to have been any
meeting of the minds previously on whether those should match or not,
but at least some people have been trying to make 'em match.
Jesse Zhang, Tom Lane
Discussion: https://postgr.es/m/CAGf+fX7DDyq6WfCy6X_KtD28MkbNBE6NkRi26fSf25dfUwX0zw@mail.gmail.com
This also makes an attempt to flesh out the docs for some of the more
severely underdocumented geometric operators and functions.
This effort exposed that the point <^ point (point_below) and
point >^ point (point_above) operators are misnamed; they should be
<<| and |>>, because they act like the other operators named that
way and not like the other operators named <^ and >^. But I just
documented them that way; fixing it is matter for another patch.
The haphazard datatype coverage of many of the operators is also
now depressingly obvious.
Discussion: https://postgr.es/m/158110996889.1089.4224139874633222837@wrigleys.postgresql.org
This Assert was trying to ensure that the number of columns in the foreign
key being cloned was the same number of attributes in the parentRel. Of
course, it's perfectly valid to have columns in the table which are not
part of the foreign key constraint. It appears that this Assert was
misunderstanding that.
Reported-by: Rajkumar Raghuwanshi
Reviewed-by: amul sul
Discussion: https://postgr.es/m/CAKcux6=z1dtiWw5BOpqDx-U6KTiq+zD0Y2m810zUtWL+giVXWA@mail.gmail.com
readOneRecord() is used now when looking for a checkpoint record to
check if the target server is an ancestor of the source across multiple
timelines, and using a restore_command if available improves the
stability of the operation. This part was missed in a7e8ece.
Reported-by: Kyotaro Horiguchi
Discussion: https://postgr.es/m/20200421.150830.1410714948345179794.horikyota.ntt@gmail.com
It's important to know that a trigger is cloned from a parent table,
because of the behavior that the trigger is dropped on detach. Make
psql's \d display it.
We'd like to backpatch, but lack of the pg_trigger.tgparentid column
makes it more difficult. Punt for now. If somebody wants to volunteer
an implementation that reads pg_depend on older versions, that can
probably be backpatched.
Authors: Justin Pryzby, Amit Langote, Álvaro Herrera
Discussion: https://postgr.es/m/20200419002206.GM26953@telsasoft.com
Checking if Subject Alternative Names (SANs) from a certificate match
with the hostname connected to leaked memory after each lookup done.
This is broken since acd08d7 that added support for SANs in SSL
certificates, so backpatch down to 9.5.
Author: Roman Peshkurov
Reviewed-by: Hamid Akhtar, Michael Paquier, David Steele
Discussion: https://postgr.es/m/CALLDf-pZ-E3mjxd5=bnHsDu9zHEOnpgPgdnO84E2RuwMCjjyPw@mail.gmail.com
Backpatch-through: 9.5
Add a couple of lines to make it explicit that indexes, constraints,
triggers are added, removed, or left alone.
Backpatch to pg11.
Author: Álvaro Herrera <alvherre@alvh.no-ip.org>
Reviewed-by: Justin Pryzby <pryzby@telsasoft.com>
Discussion: https://postgr.es/m/20200421162038.GA18628@alvherre.pgsql
index.c supposed that it could just use a PG_TRY block to clean up the
state associated with an active REINDEX operation. However, that code
doesn't run if we do a FATAL exit --- for example, due to a SIGTERM
shutdown signal --- while the REINDEX is happening. And that state does
get consulted during catalog accesses, which makes it problematic if we
do any catalog accesses during shutdown --- for example, to clean up any
temp tables created in the session.
If this combination of circumstances occurred, we could find ourselves
trying to access already-freed memory. In debug builds that'd fairly
reliably cause an assertion failure. In production we might often
get away with it, but with some bad luck it could cause a core dump.
Another possible bad outcome is an erroneous conclusion that an
index-to-be-accessed is being reindexed; but it looks like that would
be unlikely to have any consequences worse than failing to drop temp
tables right away. (They'd still get dropped by the next session that
uses that temp schema.)
To fix, get rid of the use of PG_TRY here, and instead hook into
the transaction abort mechanisms to clean up reindex state.
Per bug #16378 from Alexander Lakhin. This has been wrong for a
very long time, so back-patch to all supported branches.
Discussion: https://postgr.es/m/16378-7a70ca41b3ec2009@postgresql.org
Working on commit 1c455078b led me to check through FunctionCallInvoke
call sites to see if every one was being honest about (a) making sure
that fcinfo.isnull is initially false, and (b) checking its state after
the call. Sure enough, I found some violations.
The main one is that finalize_partialaggregate re-used serialfn_fcinfo
without resetting isnull, even though it clearly intends to cater for
serialfns that return NULL. There would only be an issue with a
non-strict serialfn, since it's unlikely that a serialfn would return
NULL for non-null input. We have no non-strict serialfns in core, and
there may be none in the wild either, which would account for the lack
of complaints. Still, it's clearly wrong, so back-patch that fix to
9.6 where finalize_partialaggregate was introduced.
Also, arrayfuncs.c and rowtypes.c contained various callers that were
not bothering to check for result nulls. While what's being called is
a comparison or hash function that probably *shouldn't* return null,
that's a lousy excuse for not having any check at all. There are
existing places that just Assert(!fcinfo->isnull) in comparable
situations, so I added that to the places that were calling btree
comparison or hash support functions. In the places calling
boolean-returning equality functions, it's quite cheap to have them
treat isnull as FALSE, so make those places do that. Also remove some
"locfcinfo->isnull = false" assignments that are unnecessary given the
assumption that no previous call returned null. These changes seem like
mostly neatnik-ism or debugging support, so I didn't back-patch.
When a partition is detached, any triggers that had been cloned from its
parent were not properly disentangled from its parent triggers.
This resulted in triggers that could not be dropped because they
depended on the trigger in the trigger in the no-longer-parent table:
ALTER TABLE t DETACH PARTITION t1;
DROP TRIGGER trig ON t1;
ERROR: cannot drop trigger trig on table t1 because trigger trig on table t requires it
HINT: You can drop trigger trig on table t instead.
Moreover the table can no longer be re-attached to its parent, because
the trigger name is already taken:
ALTER TABLE t ATTACH PARTITION t1 FOR VALUES FROM (1)TO(2);
ERROR: trigger "trig" for relation "t1" already exists
The former is a bug introduced in commit 86f575948c. (The latter is
not necessarily a bug, but it makes the bug more uncomfortable.)
To avoid the complexity that would be needed to tell whether the trigger
has a local definition that has to be merged with the one coming from
the parent table, establish the behavior that the trigger is removed
when the table is detached.
Backpatch to pg11.
Author: Justin Pryzby <pryzby@telsasoft.com>
Reviewed-by: Amit Langote <amitlangote09@gmail.com>
Reviewed-by: Álvaro Herrera <alvherre@alvh.no-ip.org>
Discussion: https://www.postgresql.org/message-id/flat/20200408152412.GZ2228@telsasoft.com
Commit 0d861bbb, which introduced deduplication to nbtree, added some
logic to take large posting list tuples into account when choosing a
split point. We subtract firstright posting list overhead from the
projected new high key size when calculating leftfree/rightfree values
for an affected candidate split point. Posting list tuples aren't
special to nbtsplitloc.c, but taking them into account like this makes a
huge difference in practice. Posting list tuples are frequently tuple
size outliers.
However, commit 0d861bbb missed a closely related issue: split interval
itself is calculated based on the assumption that tuples on the page
being split are roughly equisized. That assumption was acceptable back
when commit fab25024 taught the logic for choosing a split point about
suffix truncation, but it's pretty questionable now that very large
tuple sizes are common. This oversight led to unbalanced page splits in
low cardinality multi-column indexes when deduplication was used: page
splits that don't give sufficient weight to how unbalanced the split is
when the interval happens to include some large posting list tuples (and
when most other tuples on the page are not so large).
Nail this down by calculating an initial split interval in a way that's
attuned to the actual cost that we want to keep under control (not a
fuzzy proxy for the cost): apply a leftfree + rightfree evenness test to
each candidate split point that actually gets included in the split
interval (for the default strategy). This replaces logic that used a
percentage of all legal split points for the page as the basis of the
initial split interval.
Discussion: https://postgr.es/m/CAH2-WznJt5aT2uUB2Bs+JBLdwe0XTX67+xeLFcaNvCKxO=QBVQ@mail.gmail.com
Although selfuncs.c will never call a target operator with null inputs,
some functions might return null anyway. The existing coding will fail
if that happens (since FunctionCall2Coll will punt), which seems
undesirable given that matchingsel() has such a broad range of potential
applicability --- in fact, we already have a problem because we apply it
to jsonb_path_exists_opr, which can return null. Hence, rejigger the
underlying functions mcv_selectivity and histogram_selectivity to cope,
treating a null result as false.
While we are at it, we can move the InitFunctionCallInfoData overhead
out of the inner loops, which isn't a huge number of cycles but might
save something considering we are likely calling functions as cheap
as int4eq(). Plus, the number of loop cycles to be expected is much
more than it was when this code was written, since typical settings
of default_statistics_target are higher.
In view of that consideration, let's apply the same change to
var_eq_const, eqjoinsel_inner, and eqjoinsel_semi. We do not expect
equality functions to ever return null for non-null inputs (and
certainly that code has been that way a long time without complaints),
but the cycle savings seem attractive, especially in the eqjoinsel loops
where there's potentially an O(N^2) savings.
Similar code exists in ineq_histogram_selectivity and
get_variable_range, but I forebore from changing those for now.
The performance argument for changing ineq_histogram_selectivity
is really weak anyway, since that will only iterate log2(N) times.
Nikita Glukhov and Tom Lane
Discussion: https://postgr.es/m/9d3b0959-95d6-c37e-2c0b-287bcfe5c705@postgrespro.ru
Older gcc versions don't like duplicate typedefs, so get rid of
that in favor of doing it like we do it elsewhere, ie just use
a "struct" declaration when trying to avoid importing a whole
header file.
Also, there seems no reason to include stringinfo.h here at all,
so get rid of that addition too.
Discussion: https://postgr.es/m/27239.1587415696@sss.pgh.pa.us
Previously in the "Standby Server Operation" section, pg_ctl promote and
protmote_trigger_file were documented as a method to trigger standby
promotion, but pg_promote() function not.
This commit also adds parentheses into <function>pg_promote</function>
in some docs to make it clearer that a function is being referred to.
Author: Masahiro Ikeda
Reviewed-by: Michael Paquier, Laurenz Albe, Tom Lane, Fujii Masao
Discussion: https://postgr.es/m/de0068417a9f4046bac693cbcc00bdc9@oss.nttdata.com
... as added in the prior commit.
(We'd like to have tab-completion for the other object types too, but
they don't have sub-command completion yet.)
Author: Ibrar Ahmed <ibrar.ahmad@gmail.com>
Reviewed-by: Álvaro Herrera <alvherre@alvh.no-ip.org>
Discussion: https://postgr.es/m/CALtqXTcogrFEVP9uou5vFtnGsn+vHZUu9+9a0inarfYVOHScYQ@mail.gmail.com
Commit f2fcad27d5 (9.6 era) added the ability to mark objects as
dependent an extension, but forgot to add a way for such dependencies to
be removed. This commit fixes that oversight.
Strictly speaking this should be backpatched to 9.6, but due to lack of
demand we're not doing so at this time.
Discussion: https://postgr.es/m/20200217225333.GA30974@alvherre.pgsql
Reviewed-by: ahsan hadi <ahsan.hadi@gmail.com>
Reviewed-by: Ibrar Ahmed <ibrar.ahmad@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Along the way, update the older examples for bytea to use "hex"
output format. That lets us get rid of the lame disclaimer about
how the examples assume bytea_output = escape, which was only half
true anyway because none of the more-recently-added examples had
paid any attention to that.
The views pg_stat_progress_* had not gotten the memo that
pg_read_all_stats is supposed to be able to read all statistics. Also
make a pass over all text-returning pg_stat_xyz functions that could
return "insufficient privilege" and make sure they also respect
pg_read_all_status.
Reported-by: Andrey M. Borodin
Reviewed-by: Andrey M. Borodin, Kyotaro Horiguchi
Discussion: https://postgr.es/m/13145F2F-8458-4977-9D2D-7B2E862E5722@yandex-team.ru
I took the opportunity to do some copy-editing in this area as well,
and to add some new material such as a note about BETWEEN's syntactical
peculiarities.
Of note is that quite a few of the examples of transcendental functions
needed to be updated, because the displayed output no longer matched
what you get on a modern server. I believe some of these cases are
side-effects of the new Ryu algorithm in float8out. Others appear to be
because the examples predate the addition of type numeric, and were
expecting that float8 calculations would be done although the given
syntax would actually lead to calling the numeric function nowadays.
The normalization-check target needs to be run last, after moving the
newly generated files into place. Also, we need an additional
dependency so that unicode_norm.o is rebuilt first. Otherwise,
norm_test will still test the old files but against the new expected
results, which will probably fail.
Jonathan Katz felt that slightly different indentation settings made
for a better-looking result, so sync stylesheet-fo.xsl (for PDF) and
stylesheet.css (for non-website-style HTML) with those choices.
Discussion: https://postgr.es/m/31464.1587156281@sss.pgh.pa.us
We have repeatedly seen the buildfarm reach the Assert(false) in
SyncRepGetSyncStandbysPriority. This apparently is due to failing to
consider the possibility that the sync_standby_priority values in
shared memory might be inconsistent; but they will be whenever only
some of the walsenders have updated their values after a change in
the synchronous_standby_names setting. That function is vastly too
complex for what it does, anyway, so rewriting it seems better than
trying to apply a band-aid fix.
Furthermore, the API of SyncRepGetSyncStandbys is broken by design:
it returns a list of WalSnd array indexes, but there is nothing
guaranteeing that the contents of the WalSnd array remain stable.
Thus, if some walsender exits and then a new walsender process
takes over that WalSnd array slot, a caller might make use of
WAL position data that it should not, potentially leading to
incorrect decisions about whether to release transactions that
are waiting for synchronous commit.
To fix, replace SyncRepGetSyncStandbys with a new function
SyncRepGetCandidateStandbys that copies all the required data
from shared memory while holding the relevant mutexes. If the
associated walsender process then exits, this data is still safe to
make release decisions with, since we know that that much WAL *was*
sent to a valid standby server. This incidentally means that we no
longer need to treat sync_standby_priority as protected by the
SyncRepLock rather than the per-walsender mutex.
SyncRepGetSyncStandbys is no longer used by the core code, so remove
it entirely in HEAD. However, it seems possible that external code is
relying on that function, so do not remove it from the back branches.
Instead, just remove the known-incorrect Assert. When the bug occurs,
the function will return a too-short list, which callers should treat
as meaning there are not enough sync standbys, which seems like a
reasonably safe fallback until the inconsistent state is resolved.
Moreover it's bug-compatible with what has been happening in non-assert
builds. We cannot do anything about the walsender-replacement race
condition without an API/ABI break.
The bogus assertion exists back to 9.6, but 9.6 is sufficiently
different from the later branches that the patch doesn't apply at all.
I chose to just remove the bogus assertion in 9.6, feeling that the
probability of a bad outcome from the walsender-replacement race
condition is too low to justify rewriting the whole patch for 9.6.
Discussion: https://postgr.es/m/21519.1585272409@sss.pgh.pa.us
In some corner cases, this could also lead to corrupted values being
included in the tuple.
Users who are concerned that they are affected by this should first
upgrade and then perform a base backup of their database and restore onto
an off-line server. They should then query each table with generated
columns to ensure there are no rows where the generated expression does
not match a newly calculated version of the GENERATED ALWAYS expression.
If no crashes occur and no rows are returned then you're not affected.
Fixes bug #16369.
Reported-by: Cameron Ezell
Discussion: https://postgr.es/m/16369-5845a6f1bef59884@postgresql.org
Backpatch-through: 12 (where GENERATED ALWAYS columns were added.)
The table layout ideas proposed in commit e894c6183 were not as widely
popular as I'd hoped. After discussion, we've settled on a layout
that's effectively a single-column table with cell contents much like a
<varlistentry> description of the function or operator; though we're not
actually using <varlistentry>, because it'd add way too much vertical
space. Instead the effect is accomplished using line-break processing
instructions to separate the description and example(s), plus CSS or FO
customizations to produce indentation of all but the first line in each
cell. While technically this is a bit grotty, it does have the
advantage that we won't need to write nearly as much boilerplate markup.
This patch updates tables 9.30, 9.31, and 9.33 (which were touched by
the previous patch) to the revised style, and additionally converts
table 9.10. A lot of work still remains to do, but hopefully it won't
be too controversial.
Thanks to Andrew Dunstan, Pierre Giraud, Robert Haas, Alvaro Herrera,
David Johnston, Jonathan Katz, Isaac Morland for valuable ideas.
Discussion: https://postgr.es/m/8691.1586798003@sss.pgh.pa.us
Tom Lane